Knowledge IBM Tuning Statistics

Below are the keys to interpreting the different parameter ‘levels’ of each Scenario. We have a total of 12,500 unique Scenarios

Vital Rates, K Selected

age	birthRate	survivalRate	N0_proportion
0	0.00	0.70	0.5181199
1	1.40	0.95	0.2403399
2	1.90	0.90	0.1513028
3	1.75	0.00	0.0902375

Vital Rates, R Selected

age	birthRate	survivalRate	N0_proportion
0	0	0.14	0.9041858
1	14	0.18	0.0842448
2	16	0.22	0.0100919
3	12	0.00	0.0014776

Boldness rbeta Shape Parameters

Parameter	Shape	skewLow	skewMedLow	uniform	skewMedHigh	skewHigh
Boldness	rbeta: Shape1	1	2	1	5	5
Boldness	rbeta: Shape2	5	5	1	2	1

Poisson Lambda, Information Transfer, and Naive Learning

Parameter	Low	MedLow	Med	MedHigh	High
rpois: Lambda	0.000	1.000	2.00	3.0	4.0
probability: InfoTransfer	0.005	0.010	0.05	0.1	0.5
probability: InfoDeath	0.010	0.100	0.20	0.4	0.6
rbinom: naiveLearn	0.000	0.005	0.01	0.1	0.5

Vertical Transmission

Parameter	True	False
Vertical Transmission	TRUE	FALSE

---

Boxplots on the general effects of each parameter

The x-axis labels follow this pattern: ‘Vitals - End Result - Parameter Level - Mean Parameter Value (for survived and extirpated results)’. This is across all scenarios, with each scenario having 50 iterations. So these data are grouped to a certain extent to provide you with general patterns.

Attempt to try and quantify the results of the above plots

By taking the same data the the Box Plots are based off of, I fit individual linear models (ANOVA) to each dataset, in the for of MeanYears ~ Scenario. I then captured the Estimates for each Scenario into a table and each column is the model parameter estimate for the scenario given that parameter. This was the first method that popped into my head to try and compare effect sizes between different IBM parameters. I then performed a complete Pairwise Contrast between all scenarios within a parameter model. The scatter plots below are some graphical representations of the covariation between two IBM parameters in Effect-size Space. I hope this makes sense and is somewhat useful. All parameters seems to covary positively with the others, but there are some patterns when you look across the K vs R selected space and whether the model populations survived or were extirpated. The plots are dynamic, so you can turn on/off each point by double-clicking in the legend and hover allows you to see values.

Here are the actual tables of the Effect Sizes for the linear models

This table represent the Effect size of each Scenario on MeanYears to reach K or Crash. I did not calculate any differences between effect sizes, but you could do so if you feel it would be informative.

This table represent the Effect size Pairwise comparisons of each Scenario on MeanYears to reach K or Crash. This is the table used for the above Scatterplots. FYI: For this to work, I renamed the Bold categories to be Low - High, so I removed the prefix ‘skew’, and I also renamed ‘uniform’ to = ‘Med’. Just to allow for easier data joining.